thinkhard.SI

Prompt repository Assistants & system prompts

Write instructions for a Custom GPT, Claude Project or Gem

You need a reusable assistant for a repeated task and want clear instructions and a list of files to attach.

Registry no.
SYS-007
Collection
Assistants & system prompts
Laid out for
Any chat model
Score
10 out of 10
Blanks to fill
7
Added

Assistants & system promptsAny chat

The prompt

Score 10 of 10, Excellent

Use it when You need a reusable assistant for a repeated task and want clear instructions and a list of files to attach.

  • {{platform}}
  • {{task}}
  • {{users}}
  • {{inputs}}
  • {{rules}}
  • {{available_files}}
  • {{good_example}}
Full prompt and why it works
Prompt
Help me set up a reusable assistant in {{platform}} for a task I do often. I need the instructions to paste in and a list of the files to attach, with a clear reason for what goes where.

The task: {{task}}
Who will use the assistant: {{users}}
What people will paste or type in each time: {{inputs}}
Rules it must follow and things it must never do: {{rules}}
Files I could attach (style guides, templates, policies, past examples): {{available_files}}

An example of a really good result for this task:
"""
{{good_example}}
"""

How to divide the work:
- Instructions hold what applies to every conversation: the role, who the users are, the steps to follow, the output format, the rules, and what to do when information is missing. Keep every essential rule here, not only in a file.
- Files hold long or changing reference material, such as policies, product facts, price lists, templates and examples. Assistants may search files for relevant passages rather than reading every word, so the instructions should name each file and say when to use it, for example "For prices, use the price list file only. If a price is not there, say so."
- Leave out anything confidential. People who use a shared assistant can often get it to reveal its instructions and file contents.

Produce:
1. A name and a one-line description for the assistant.
2. The instructions, ready to paste, in plain sections: Role, What users bring, Steps, Output format, Rules, When information is missing. Aim for 300 to 800 words; shorter is better as long as nothing essential is lost. They should be clear enough that a colleague reading them could predict what the assistant will do.
3. Files to attach: a table with the columns File, What it contains, Why it is a file and not an instruction, and How to prepare it, for example removing personal data or keeping only the current version.
4. Settings: if {{platform}} lets me switch tools such as web search or file creation on or off, which ones this assistant needs and why.
5. Four example first messages a user might send.
6. Five test requests, each with what a correct response does, including one with missing information and one that asks the assistant to break a rule.

Use only the facts I gave, and mark anything you had to assume with "[assumed: ...]". If the task is too vague to write the steps, ask me up to three questions before drafting.

Variables

{{platform}}
Where you are building it, for example "a Custom GPT in ChatGPT", "a Claude Project" or "a Gemini Gem".
{{task}}
The job the assistant does, for example "turn rough notes into client proposals in our format".
{{users}}
For example "just me" or "our five-person support team".
{{inputs}}
What each conversation starts with, for example "a customer email" or "rough meeting notes".
{{rules}}
What it must always do and never do, for example "never quote prices that are not in the price list".
{{available_files}}
The documents you have, one per line, with a few words on each.
{{good_example}}
An output you were happy with, to show the standard. Remove anything confidential first.

Why it works

  • Background facts. Rules that apply every time go in the instructions and long reference material goes in files the instructions name, which keeps key rules from getting lost in a document.
  • What to avoid. It warns you to leave confidential material out, because people using a shared assistant can often get it to show its instructions and files.
  • An example. Your own example of a good result sets the standard, and five test requests let you check the assistant before anyone else relies on it.

How to use it

  1. Press Fill in variables and type your details into each blank. What you type stays in this browser.
  2. Copy the filled prompt into the AI tool you already use, such as ChatGPT, Claude or Gemini, or open it in the Grader or the Prompt Tester here.
  3. Before you rely on it, try it on three or four real examples. The Prompt Tester runs them all at once and shows which ones go wrong.
Use it from code

Get this prompt as JSON from the free REST API, with your own text in the blanks, or let an AI assistant fetch it through the MCP server. No key.

curl -s https://thinkhard.si/api/v1/prompts/custom-assistant-instructions

With the MCP server added to Claude, ChatGPT, Cursor or VS Code, call get_prompt with {"prompt": "custom-assistant-instructions"}, or just ask for it by name.